Tobacco Yield Estimation Based On UAV Multi-spectral Remote Sensing Image

Taking the tobacco field production demonstration area of Shixing County(North Guangdong)in Shaoguan City of Guangdong Province as the research area,the remote sensing images at the phases of tobacco root extension(April 21),the prosperous(May 19)and the mature(June 2)were acquired by the domestically made DJI drone Phantom 4 Pro equipped with Parrot Sequoia multispectral camera to estimate tobacco yield.According to tobacco multi-spectral images of different phases,the vegetation index(VI)was extracted,and the correlation between yield and VI was analyzed.The VI with correlation P<0.05 was selected as the explanatory variable,and the measured yield was selected as the dependent variable.Tobacco yield estimation models of linear and curvilinear were constructed based on the multiple time phases of tobacco,and the accuracy of the models was verified.The main conclusions of this study are as follows:(1)The correlation coefficients between VI including NDVI,MTVI1,MSR,GNDVI and EVI2 and tobacco yield at each phase were all above 0.5,and they could be selected as the predicting factors for tobacco production estimation.(2)The maturity phase is the best period to estimate the yield of tobacco.Through comparative analysis,the yield estimation model constructed by the MSR has the greatest accuracy at this period.In the single VI linear yield estimation model,the yield-MSR estimation model R2 was 0.642,the relative error RE was 1.86,and the root mean square error RMSE was 33.87;among the nonlinear yield estimation models,the yield-MSR polynomial yield estimation model R2 has the greatest accuracy of 0.695,which were 1.38% and 31.27 for RE and RMSE,respectively.Therefore,to estimate yield based on the tobacco canopy VI,a polynomial nonlinear yield estimation model is the best choice,and the best yield estimation VI is MSR,and the best yield estimation period is the mature period.(3)Comparing the coefficient of determination of the single VI model,it can be found that the accuracy of the linear regression yield estimation model is lower than that of the nonlinear regression yield estimation model,which is consistent with the result of the VI combination yield estimation model.Among the estimation models of combined VI,the polynomial non-linear yield estimation model of yield-MSR-GNDVI has the greatest accuracy,which were the multiple VI and multiple phases combination(high season and mature period).The best tobacco yield estimation model has the greatest constructing accuracy,which was R2 of 0.795,RE of 1.91%,and RMSE of 25.64,and the greatest verifying accuracy,which was R2 of 0.729,RMSE of 29.59,and RE of 2.06%.